Biomedical Engineering Reference
In-Depth Information
performance of any model is directly affected by the training set used in its
development, the training set should fulfil certain criteria. It should:
• Contain all expected components
• Span the concentration ranges of interest
• Span the conditions of interest
• Contain mutually independent samples.
The calibration should also be validated using a set of samples (validation set)
which is independent of the training set. Strategies on how to determine an
experimental design which will achieve these aims can be found elsewhere
[
123
,
127
,
128
].
Partial Least-Squares Regression
An often-used chemometric calibration technique for bioprocessing applications is
partial least-squares regression (PLS). This is a multivariate statistical technique
developed from classical least-squares and inverse least-squares regression used in
economic forecasting and later in chemical applications [
127
] (see Chap. 7).
2.3.5 PAT Applications of Vibrational Spectroscopy in Bioprocessing
The applications or potential applications of vibrational spectroscopy in biopro-
cessing are largely dependent on the sampling interfaces available. A number of
instruments exist, and sample interfaces vary from sample cavities using cuvettes
or vials to immersion probes. Where real-time data are required for monitoring and
control purposes, the type of available instruments is very much reduced, as all
offline techniques are eliminated. Bioprocess applications to date have used either
flow cells, where the sample of interest is passed through a measuring chamber, or
immersion probes, where a probe is inserted into a reactor and the sample is
scanned in situ by transflectance, transmission or reflectance methods. The
development of high-quality fibre optics and autoclavable probes has increased the
capabilities of these techniques. The most common applications in bioprocessing
are analyte, metabolite and biomass monitoring, with monitoring systems in some
cases further developed to enable process control.
MIR Applications
MIR lags behind its infrared counterpart, NIR, when it comes to applications in
bioprocessing. Despite the fact that MIR can detect and quantify components in
aqueous solutions at significantly lower levels than NIR [
121
], MIR is less
extensively used. Only in the last decade has MIR been considered a potentially
useful tool for bioprocess monitoring. Work to date has mainly focussed on
Search WWH ::
Custom Search